低速率LD-CELP 语音编码算法研究

This research purpose is design and present a high quality (MOS4.0) speech coding algorithm with low code rate (8Kb/s) and low delay (less than 2.5ms), which is a challenge task in communication field. The key-point solving this problem depends on designing and exploring a.method of exact denoting exciting gain then training further normal hape-codebook based on it.The fundamental breakthrough has been made out. With the scheme of adaptive quantization gain provided by this research combining the Jayant adaptive quantization and G.728’s gain prediction,for adaptive quantization gain of 3 bits and 4 bits, the SNR have been increased separately by 0.5and 6 dB than that of G.728. If adopting 4 bits adaptive quantization and the shape codebook of 64.codevectors, the G.728’ complexity could be reduced by 20% while SNR had 1 dB improvement.There is important trait of this result that the gain of exciting-vector of 5 to 20 samples can be.exactly quantized only with 2 or 3 bits (no sign bit), and the SNR is over 17 dB. Since training normal shape-codebook has nothing out of the ordinary, this result is really the fundament finding.out various speech-coding algorithm delaying from 5 to 20 samples.gain adaptive quantization;

申请者前期研究表明，采用归一化波形码书并精确表示激励增益是改善低延迟语音编码效果的极好途径。本研究将提出延迟1ms，长话音质的8kb/s语音编码方案。此项成果可实现电信网设备8倍扩容并与移动网和IP网语音实现无译码转接。研究涉及新的思路如波形码书与增媛瞬ㄆ鞯墓叵担煌ü繁硎炯だ鲆娲婊粼げ獾龋哉庖涣煊蛴凶胖匾睦砺垡庖濉